Grouping images into semantically meaningful categories using low-level visual feature is a challenging and important problem in content-based image retrieval. The groupings can be used to build effective indices for an image database. Digital image analysis techniques are being used widely in remote sensing assuming that each terrain surface category is characterized with spectral signature observed by remote sensors. Even with the remote sensing images of IRS data, integration of spatial information is expected to assist and to improve the image analysis of remote sensing data. In this paper we present a satellite image retrieval based on a mixture of old fashioned ideas and state of the art learning tools. We have developed a methodology to classify remote sensing images using HSV color features and Haar wavelet texture features and then grouping them on the basis of particular threshold value. The experimental results indicate that the use of color and texture feature extraction is very useful for image retrieval.
The advent of Digital photography, reduction in cost of mass storage device and use of high-capacity public networks have led to a rapid increase in the use of digital images in various domains such as publishing, media, military and education. The need to store, manage and locate these images has become a challenging task. Generally, there are two main approaches for classifying images: image classification based on keywords and the other one is content based image retrieval. The former technique suffers from the need for manual classification of images, which is simply not practical in a large collection of image. Further incompleteness of a limited set of keyword descriptors may significantly reduce query effectiveness at the time of image retrieval. In latter technique images can be identified by automatic description, which depends on their objective visual content.
Remote Sensing Application images are depicted using spatial distribution of a certain field parameters such as reflectivity of (EM) radiation, emissivity, temperature or some geophysical or topographical elevation. We have designed a system to retrieve similar remote sensing images using some traditional and modern approach.
Content Based Image Retrieval is a set of techniques for retrieving semantically relevant images from an image database based on automatically derived image features [1]. The computer must be able to retrieve images from a database without any human assumption on specific domain (such as texture vs. non texture or indoor vs. outdoor).
One of the main tasks for CBIR systems is similarity comparison, extracting feature signatures of every image based on its pixel values and defining rules for comparing images. These features become the image representation for measuring similarity with other images in the database. To compare images the difference of the feature components is calculated.
Early CBIR methods used global feature extraction to obtain the image descriptors. For example, QBIC [2], developed at the IBM Almaden Research Center, extracts several features from each image, namely color, texture and shape features. These descriptors are obtained globally by extracting information on the means of color histograms for color features; global texture information on coarseness, contrast, and direction; and shape features about the curvature, moments invariants, circularity, and eccentricity. Similarly, the Photo-book-system [3], Visual-Seek [4], and Virage [5], use global features to represent image semantics.
The system in [6] attempt to overcome previous method limitations of global based retrieval systems by representing images as collections of regions that may correspond to objects such as flowers, trees, skies, and mountains. This system applies image segmentation [7] to decompose an image into regions, which correspond to physical objects (trees, people, cars, flowers) if the decomposition is ideal. The feature descriptors are extracted on each object instead of global image. Color and texture features are extracted on each pixel that belongs to the object, and each object is described by the average value of these pixel features. In this paper color and texture feature extraction, clustering and similarity matching is used.
A system is developed for image retrieval. In this an image database of LISS III sensor is used. LISS III has a spatial resolution of 23m and a swath width of 140 km. Then the query image is taken and images similar to the query images are found on the basis of colour and texture similarity. The three main tasks of the system are:
- Colour & Texture Feature Extraction. 2. K-means clustering to form groups. 3. Similarity distance computation between the query image and database images.
We have used the approach of Li and Wang [1] and Zhang [9]. The image is partitioned into 4 by 4 blocks, a size that provides a compromise between texture granularity, segmentation coarseness, and computation time. As part of pre-processing, each 4x4 block is replaced by a single block containing the average value of the 4 by 4 block.
To segment an image into objects, six features are extracted from each block. Three features are color features, and the other three are texture features. The HSV color space is selected during color feature extraction due to its ability for easy transformation from RGB to HSV and vice versa. The quantization of HSV can produce a collection of colors that is also compact and complete [6]. These features are {H, S, and V} that are extracted from the RGB colour image.
To obtain the texture features, Haar wavelet transformation is used. The Haar wavelet is discontinuous and resembles a step function. It represents the energy in high frequency bands of the Haar wavelet transform. After a one-level wavelet transform, a 4 by 4 block is decomposed into four frequency bands, each band containing a 2 by 2 matrix of coefficients. Suppose the coefficients in the HL band are {c k+i,
This content is AI-processed based on open access ArXiv data.
